Crafty 25.0 Release

[bob]

Is there syzygy support? I don't see any, but I don't have that much faith in my powers of observation.

I have not looked at that as of yet. I probably will at some point, but I am not a big fan of even EGTBs nowadays since I had a chance to measure what they add to a chess program in terms of Elo (just 3-4-5 however). Answer was zilch, so I hardly ever even use them other than to make sure I haven't broken anything for those that do...

[Dr.Wael Deeb]

Great news....

What is the estimated Elo improvement over the 24 series,roughly that is
Dr.D

[APassionForCriminalJustic]

Is there syzygy support? I don't see any, but I don't have that much faith in my powers of observation.

I have not looked at that as of yet. I probably will at some point, but I am not a big fan of even EGTBs nowadays since I had a chance to measure what they add to a chess program in terms of Elo (just 3-4-5 however). Answer was zilch, so I hardly ever even use them other than to make sure I haven't broken anything for those that do...

3,4,5 may not be worth much of anything, but 6 men probing in search should certainly add 10 Elo or so... it does in Komodo and Houdini for instance. Perhaps you should consider it sometime...

[F. Bluemers]

excellent

The key is that now the individual "helper" threads do all the work, allocating a split block, copying the data from the parent, etc., rather than the parent doing it all.

Very nice , I take it the data to copy is not on the stack so the parent can always return even when the helpers are not done copying?

Best
Fonzy

[bob]

Great news....

What is the estimated Elo improvement over the 24 series,roughly that is
Dr.D

Single CPU, my testing shows +75 Elo. On multiple cores, probably better than +100 as the new SMP code is much improved...

Some have reported +100 without SMP, I can't confirm that...

[Dr.Wael Deeb]

Great news....

What is the estimated Elo improvement over the 24 series,roughly that is
Dr.D

Single CPU, my testing shows +75 Elo. On multiple cores, probably better than +100 as the new SMP code is much improved...

Some have reported +100 without SMP, I can't confirm that...

Wow,a huge leap Elo wise

Thanks Bob for such a great release and happy holidays regards,
Dr.D

[bob]

excellent

The key is that now the individual "helper" threads do all the work, allocating a split block, copying the data from the parent, etc., rather than the parent doing it all.

Very nice , I take it the data to copy is not on the stack so the parent can always return even when the helpers are not done copying?

Best
Fonzy

I have a bunch of split blocks, as opposed to a stack, for the SMP data. If a thread joins an existing split block because it sees it is the best choice, it just copies the data and goes. In fact, all threads join whatever split block they want. If there are none, they request that split blocks be produced, and they still join whichever one they want. Parent does no locking to split period. A thread locks for a microsecond or so to link its own split block to the parent, but the lock is then released and there is no further locking except when extracting a move from the shared move list.. Is actually very simple code now...

And yes, the parent can abandon the split block while the child is copying, although I have not caught that happening yet during testing. The only odd thing I allow is that a thread can do a "gratuitous split" at good points in the search so that others can join there when they become idle. And interestingly THIS thread can also "join itself" if it runs out of work to do after having done a split deeper in the tree. And it actually helps performance.

[bob]

Great news....

What is the estimated Elo improvement over the 24 series,roughly that is
Dr.D

Single CPU, my testing shows +75 Elo. On multiple cores, probably better than +100 as the new SMP code is much improved...

Some have reported +100 without SMP, I can't confirm that...

Wow,a huge leap Elo wise

Thanks Bob for such a great release and happy holidays regards,
Dr.D

Not so huge when you consider how long it has been since 24.1 was released. Close to 15 months...

[supersharp77]

Great news....

What is the estimated Elo improvement over the 24 series,roughly that is
Dr.D

Single CPU, my testing shows +75 Elo. On multiple cores, probably better than +100 as the new SMP code is much improved...

Some have reported +100 without SMP, I can't confirm that...

Wow,a huge leap Elo wise

Thanks Bob for such a great release and happy holidays regards,
Dr.D

Not so huge when you consider how long it has been since 24.1 was released. Close to 15 months...

Thanks Dr Hyatt!! Great News!! Crafty 25.0......Back in Business!!

[Dr.Wael Deeb]

And the binaries